Discovering BPMN Models with Sub-processes and Multi-instance Markers

Abstract

Massive event logs are produced in information systems, which record executions of business processes in organizations. Various techniques are proposed to discover process models reflecting real-life behaviors from these logs. However, the discovered models are mostly in Petri nets rather than BPMN models, the current industrial process modeling standard. Conforti et al. and Weber et al. propose techniques that discover BPMN models with sub-processes, multi-instance, etc. However, these techniques are made for event logs with special attributes, e.g., containing attributes about primary and foreign keys, which may not commonly appear in event logs. For example, logs from the OA office automation systems of CMCC China Mobile Communications Corporation do not contain such data. To solve this issue, this paper proposes two techniques that can discover BPMN models with sub-processes and multi-instance markers with event logs containing less event attributes. One of our techniques only requires four event attributes: case id, task name, start time and end time. Experimental evaluations with both real-life logs and synthetic logs show that our techniques can indeed discover process models with sub-process and multi-instance markers from logs with less event attributes, and are more accurate and less complex than those derived with flat process model discovery techniques.